The extensive work on transfer of analogical training across domains has resulted in limited success. One line of research that has shown promise is training on statistical concepts such as the law of large numbers. This dissertation seeks to expand on the promise of that line of inquiry, demonstrating that training participants on fast and frugal statistical heuristics (i) improves prospective conditional and counterfactual forecast accuracy and (ii) shows the promise of transfer across domains. In this dissertation we present four studies (Total N=1,541) that demonstrate the effectiveness of training on statistical concepts to improve accuracy of counterfactual forecasts and prospective conditional forecasts. Training consisted of a series of videos on fast and frugal statistical heuristics that taught participants to (i) calculate base rates and start their forecasts there, (ii) identify player patterns, and (iii) forecast away from the base rates only if the counterfactual change significantly alters the normal pattern of play. We asked novice participants to make forecasts about the outcome of simulated runs of the pure-strategy card game Goofspiel. This paradigm allowed assessment of whether training could be effective in overcoming two automatic cognitive processes—the anchoring heuristic and retrospective determinism—that can lock us into seeing our shared single run of history as the natural (meant-to-be) state-of-affairs. We find that training novice forecasters on fast and frugal statistical heuristics resulted in substantial improvements in the accuracy of their counterfactual forecasts within the training domain (Studies 1, 2, and 3). We also find that these fast and frugal statistical heuristics show promise for limited transportability across domains (Study 4). Two lines of follow-up research look promising: (1) assessing the transportability of counterfactual forecasting skills across a range of simulated worlds and inferential tasks, including forward-in-time forecasting and decision making; and (2) assessing whether techniques that improve accuracy in simulations also improve real-world judgments, with emphasis on exploring the links between retrospective and prospective what-if judgments.